Free tool · No sign-up required

Agent Trace Inspector

Paste any OTLP JSON trace from LangSmith, Langfuse, or the OpenTelemetry Collector and instantly visualize the agent graph. Reads gen_ai.* semantic conventions — no SDK changes required.

Try Agent Trace Inspector →Free to use

Frameworks, exporters, and conventions detected

LangGraphCrewAIAutoGenOpenAI Agents SDKLangSmithLangfuseOpenTelemetrygen_ai.agent.namegen_ai.tool.namegen_ai.usage.input_tokensgen_ai.usage.output_tokensgen_ai.systemOTLP JSONGPT-4oClaude 3.5 SonnetGemini 1.5 Pro

Features

From raw trace to agent graph in seconds

Instant agent graph

Paste any OTLP JSON trace and the tool renders the agent-tool graph in milliseconds. No SDK installation, no local collector, no live traffic needed.

gen_ai semantic conventions

Reads the OpenTelemetry GenAI semantic conventions — gen_ai.agent.name, gen_ai.tool.name, gen_ai.request.model, gen_ai.system — from every span to classify nodes without heuristics.

Token attribution

LLM call spans without an agent name (raw chat spans) are automatically folded onto their nearest ancestor agent node. Per-agent token totals reflect every inference call that agent triggered.

Framework auto-detection

The gen_ai.system attribute is used to detect LangGraph, CrewAI, AutoGen, and OpenAI Agents SDK automatically. Handoff edges (agent-to-agent) are rendered with animated purple arrows.

Span timeline

A waterfall view of every span, sorted by start time, with duration bars and per-span token counts visible at a glance. Pro feature.

Works with any OTLP exporter

Accepts traces exported from LangSmith, Langfuse, Jaeger, Zipkin, Grafana Tempo, or the OpenTelemetry Collector. Any service exporting the GenAI semantic conventions is supported.

How span attributes map to graph nodes

gen_ai.agent.name = "ResearchAgent"→Agent node

gen_ai.tool.name = "web_search"→Tool node

gen_ai.operation.name = "chat" (no agent name)→Folded onto parent agent

parentSpanId: null→Entry (START) node

agent with no child agent span→Exit (END) node

Who uses it

AI engineers

Debug LangGraph ReAct loops, CrewAI pipeline handoffs, and AutoGen group chat flows by inspecting the exact span tree from a failed production run.

Platform / observability teams

Validate that every agent, tool, and LLM call in your multi-agent system is properly instrumented before a new service goes to production.

Security engineers

Audit the complete LLM call chain for a given user session — which agents ran, which tools were invoked, what models were called, and how many tokens each consumed.

Try Agent Trace Inspector free →
See all ContextIQ tools